Ashutosh Kushwaha

Designed a multi-stage propensity modeling framework with funnel-aware feature engineering and stage-wise intent prediction for Insurance, Investments, and Credit Cards, enabling precision targeting for 50M+ users and improving funnel conversion by 19 BPS.

Operationalized weekly scoring pipelines for propensity models (stacked ensemble of XGBoost + DNN) and integrated outputs with the outbound CRM platform/call center to enable automated, timely targeting of high-intent users.

Led end-to-end development of a zone-aware trend discovery and product ranking system for Tata CLiQ merchandising, partnering with product and engineering to define requirements, productionize models, and run live A/B experiments, resulting in +0.34% carousel CTR uplift across regions.

Built a scalable, configurable training framework with automated large-scale feature selection (10k+ features), variance filtering, and multi-model voting-based feature subset selection.

Developed flexible Spark/AutoGluon training pipelines with integrated MLflow model registry support on Databricks, enabling automated packaging, versioning, and inference-ready deployment.

Architected and productionized a high-throughput Insurance Anti-Fraud service using FastAPI, achieving a P99 latency of 115ms for real-time screening.